05Nov

On Forecasting, Big Data and Parsimony

Author: Dan Skodol

Isaac Newton once stated, “We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances.” His quote is perhaps a modern interpretation of Occam’s razor, a problem-solving principle dating back to the 14th century. The essence of Occam’s razor is essentially that the simplest solution is usually the correct one.

graph of data points against an electronic grid

Fast-forward to the 21st century, and enter “Big Data.” I’ll use this term loosely, as I’ve seen numerous definitions and uses of the phrase recently, but for the best definition see here. Regardless, it seems the proliferation of new and diverse data sources have lit a bit of a fire underneath those in the revenue management space who are perpetually in search of the “holy grail” of forecast accuracy, as it’s generally proven that the more accurate a forecast, the greater the revenue opportunity to be captured by the firm employing it. If inclusion of this new data can aid forecast accuracy, then certainly we should be building it into our models – right?

Forecasting Hotel Bookings with Big Data

It seems rather intuitive that I should be able to derive a better forecast of my future hotel bookings if I know how much airlift is going into my market, what the weather in my area will be like, how many people are searching for my hotel on Google, and how well my reviews have been trending on social media and sites like Trip Advisor. While we at Rainmaker have seen empirically that some of this data can improve accuracy for certain hotels some of the time, let’s also intuitively take a “devil’s advocate” approach towards thinking about what this information could mean for my property or properties.

At face value, the greater the number of air passengers that are destined for my market, the more demand my market will naturally see for hotel room bookings in a given period. But as a revenue manager, I need a little bit more precision than that. Specifically when will these bookings materialize? Have they already booked? Do customers in my market necessarily book air travel at the same time as they book their hotel, or do they do it before or afterwards? And given my relative price positioning in my market when these customers book, how much of the overall market share might I expect to capture? There are likely several other factors that would need to be considered if I were to use this information to derive an accurate and reliable forecast.

Weather as a Booking Variable

The “hypothesis” here states that I can anticipate a surge in either bookings or cancellations stemming from a weather event in either my area or potentially a “feeder market” (with very few exceptions, I can’t think of too many major markets with a singular or limited set of feeder markets). Again, this makes intuitive sense, but for starters I’d be nervous about basing my forecast on a second forecast that we know is not always accurate.

Think about the implications of being wrong – if a forecasted rainstorm leads me to expect a spike in cancellations and to overbook my hotel more than usual, and that rainstorm doesn’t actually happen, I’ve put my hotel in a very difficult situation. And what of my airport hotel that does experience a significant weather event? Does that strand passengers at their origins, causing a significant last-minute drop-off in demand, or does it strand passengers at my destination, causing bookings to surge? The difference might come down to the precise timing of the weather that passes through. We’re left with the type of guesswork that might normally figure into a manual forecasting approach, and that’s not good nor is it worth my investment.

It seems these days that social media can make or break the reputation of a hotel. But what should we make of the effect of online reviews on a hotel’s demand and pricing? Research on this topic has revealed a great deal of complexity with regard to the relationships between ratings, review sentiment, price, and perceived value, such that linear relationships become difficult to establish. For instance, a hotel with positive reviews should be able to command a higher price relative to their comp set. But, when reviews go south, potential guests will avoid your hotel regardless of price, meaning it might be better to stand pat with rates and focus on correcting the root causes of your problems. Modeling the effects of reputation on demand and price elasticity for your property might prove to be a tough nut to crack.

Data, data, and more data – it may be human nature to feel that when presented with it, we need to do something with it. It’s enough to make Isaac Newton roll over in his grave. It’s best to approach this data with a careful, critical eye, taking great care to assess the value and validity of a new data source within your own forecast models. At the end of the day, the additional complexity, the corresponding investment, and the headaches that stem from encountering “dirty” data might not be worth it if they only bring about a marginal improvement to your forecasting.

Learn how Rainmaker’s revenue optimization solutions can help you

Request a Free Assessment and Get Started Today!

Recent Posts